Commit | Line | Data |
---|---|---|
10c28d93 AK |
1 | #include <linux/slab.h> |
2 | #include <linux/file.h> | |
3 | #include <linux/fdtable.h> | |
4 | #include <linux/mm.h> | |
5 | #include <linux/stat.h> | |
6 | #include <linux/fcntl.h> | |
7 | #include <linux/swap.h> | |
8 | #include <linux/string.h> | |
9 | #include <linux/init.h> | |
10 | #include <linux/pagemap.h> | |
11 | #include <linux/perf_event.h> | |
12 | #include <linux/highmem.h> | |
13 | #include <linux/spinlock.h> | |
14 | #include <linux/key.h> | |
15 | #include <linux/personality.h> | |
16 | #include <linux/binfmts.h> | |
179899fd | 17 | #include <linux/coredump.h> |
10c28d93 AK |
18 | #include <linux/utsname.h> |
19 | #include <linux/pid_namespace.h> | |
20 | #include <linux/module.h> | |
21 | #include <linux/namei.h> | |
22 | #include <linux/mount.h> | |
23 | #include <linux/security.h> | |
24 | #include <linux/syscalls.h> | |
25 | #include <linux/tsacct_kern.h> | |
26 | #include <linux/cn_proc.h> | |
27 | #include <linux/audit.h> | |
28 | #include <linux/tracehook.h> | |
29 | #include <linux/kmod.h> | |
30 | #include <linux/fsnotify.h> | |
31 | #include <linux/fs_struct.h> | |
32 | #include <linux/pipe_fs_i.h> | |
33 | #include <linux/oom.h> | |
34 | #include <linux/compat.h> | |
03927c8a | 35 | #include <linux/timekeeping.h> |
10c28d93 AK |
36 | |
37 | #include <asm/uaccess.h> | |
38 | #include <asm/mmu_context.h> | |
39 | #include <asm/tlb.h> | |
40 | #include <asm/exec.h> | |
41 | ||
42 | #include <trace/events/task.h> | |
43 | #include "internal.h" | |
44 | ||
45 | #include <trace/events/sched.h> | |
46 | ||
47 | int core_uses_pid; | |
10c28d93 | 48 | unsigned int core_pipe_limit; |
3ceadcf6 ON |
49 | char core_pattern[CORENAME_MAX_SIZE] = "core"; |
50 | static int core_name_size = CORENAME_MAX_SIZE; | |
10c28d93 AK |
51 | |
52 | struct core_name { | |
53 | char *corename; | |
54 | int used, size; | |
55 | }; | |
10c28d93 AK |
56 | |
57 | /* The maximal length of core_pattern is also specified in sysctl.c */ | |
58 | ||
3ceadcf6 | 59 | static int expand_corename(struct core_name *cn, int size) |
10c28d93 | 60 | { |
e7fd1549 | 61 | char *corename = krealloc(cn->corename, size, GFP_KERNEL); |
10c28d93 | 62 | |
e7fd1549 | 63 | if (!corename) |
10c28d93 | 64 | return -ENOMEM; |
10c28d93 | 65 | |
3ceadcf6 ON |
66 | if (size > core_name_size) /* racy but harmless */ |
67 | core_name_size = size; | |
68 | ||
69 | cn->size = ksize(corename); | |
e7fd1549 | 70 | cn->corename = corename; |
10c28d93 AK |
71 | return 0; |
72 | } | |
73 | ||
b4176b7c NI |
74 | static __printf(2, 0) int cn_vprintf(struct core_name *cn, const char *fmt, |
75 | va_list arg) | |
10c28d93 | 76 | { |
5fe9d8ca | 77 | int free, need; |
404ca80e | 78 | va_list arg_copy; |
10c28d93 | 79 | |
5fe9d8ca ON |
80 | again: |
81 | free = cn->size - cn->used; | |
404ca80e ED |
82 | |
83 | va_copy(arg_copy, arg); | |
84 | need = vsnprintf(cn->corename + cn->used, free, fmt, arg_copy); | |
85 | va_end(arg_copy); | |
86 | ||
5fe9d8ca ON |
87 | if (need < free) { |
88 | cn->used += need; | |
89 | return 0; | |
90 | } | |
10c28d93 | 91 | |
3ceadcf6 | 92 | if (!expand_corename(cn, cn->size + need - free + 1)) |
5fe9d8ca | 93 | goto again; |
10c28d93 | 94 | |
5fe9d8ca | 95 | return -ENOMEM; |
10c28d93 AK |
96 | } |
97 | ||
b4176b7c | 98 | static __printf(2, 3) int cn_printf(struct core_name *cn, const char *fmt, ...) |
bc03c691 ON |
99 | { |
100 | va_list arg; | |
101 | int ret; | |
102 | ||
103 | va_start(arg, fmt); | |
104 | ret = cn_vprintf(cn, fmt, arg); | |
105 | va_end(arg); | |
106 | ||
107 | return ret; | |
108 | } | |
109 | ||
b4176b7c NI |
110 | static __printf(2, 3) |
111 | int cn_esc_printf(struct core_name *cn, const char *fmt, ...) | |
10c28d93 | 112 | { |
923bed03 ON |
113 | int cur = cn->used; |
114 | va_list arg; | |
115 | int ret; | |
116 | ||
117 | va_start(arg, fmt); | |
118 | ret = cn_vprintf(cn, fmt, arg); | |
119 | va_end(arg); | |
120 | ||
121 | for (; cur < cn->used; ++cur) { | |
122 | if (cn->corename[cur] == '/') | |
123 | cn->corename[cur] = '!'; | |
124 | } | |
125 | return ret; | |
10c28d93 AK |
126 | } |
127 | ||
128 | static int cn_print_exe_file(struct core_name *cn) | |
129 | { | |
130 | struct file *exe_file; | |
131 | char *pathbuf, *path; | |
132 | int ret; | |
133 | ||
134 | exe_file = get_mm_exe_file(current->mm); | |
923bed03 ON |
135 | if (!exe_file) |
136 | return cn_esc_printf(cn, "%s (path unknown)", current->comm); | |
10c28d93 AK |
137 | |
138 | pathbuf = kmalloc(PATH_MAX, GFP_TEMPORARY); | |
139 | if (!pathbuf) { | |
140 | ret = -ENOMEM; | |
141 | goto put_exe_file; | |
142 | } | |
143 | ||
9bf39ab2 | 144 | path = file_path(exe_file, pathbuf, PATH_MAX); |
10c28d93 AK |
145 | if (IS_ERR(path)) { |
146 | ret = PTR_ERR(path); | |
147 | goto free_buf; | |
148 | } | |
149 | ||
923bed03 | 150 | ret = cn_esc_printf(cn, "%s", path); |
10c28d93 AK |
151 | |
152 | free_buf: | |
153 | kfree(pathbuf); | |
154 | put_exe_file: | |
155 | fput(exe_file); | |
156 | return ret; | |
157 | } | |
158 | ||
159 | /* format_corename will inspect the pattern parameter, and output a | |
160 | * name into corename, which must have space for at least | |
161 | * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator. | |
162 | */ | |
12a2b4b2 | 163 | static int format_corename(struct core_name *cn, struct coredump_params *cprm) |
10c28d93 AK |
164 | { |
165 | const struct cred *cred = current_cred(); | |
166 | const char *pat_ptr = core_pattern; | |
167 | int ispipe = (*pat_ptr == '|'); | |
168 | int pid_in_pattern = 0; | |
169 | int err = 0; | |
170 | ||
e7fd1549 | 171 | cn->used = 0; |
3ceadcf6 ON |
172 | cn->corename = NULL; |
173 | if (expand_corename(cn, core_name_size)) | |
10c28d93 | 174 | return -ENOMEM; |
888ffc59 ON |
175 | cn->corename[0] = '\0'; |
176 | ||
177 | if (ispipe) | |
178 | ++pat_ptr; | |
10c28d93 AK |
179 | |
180 | /* Repeat as long as we have more pattern to process and more output | |
181 | space */ | |
182 | while (*pat_ptr) { | |
183 | if (*pat_ptr != '%') { | |
10c28d93 AK |
184 | err = cn_printf(cn, "%c", *pat_ptr++); |
185 | } else { | |
186 | switch (*++pat_ptr) { | |
187 | /* single % at the end, drop that */ | |
188 | case 0: | |
189 | goto out; | |
190 | /* Double percent, output one percent */ | |
191 | case '%': | |
192 | err = cn_printf(cn, "%c", '%'); | |
193 | break; | |
194 | /* pid */ | |
195 | case 'p': | |
196 | pid_in_pattern = 1; | |
197 | err = cn_printf(cn, "%d", | |
198 | task_tgid_vnr(current)); | |
199 | break; | |
65aafb1e SG |
200 | /* global pid */ |
201 | case 'P': | |
202 | err = cn_printf(cn, "%d", | |
203 | task_tgid_nr(current)); | |
204 | break; | |
b03023ec ON |
205 | case 'i': |
206 | err = cn_printf(cn, "%d", | |
207 | task_pid_vnr(current)); | |
208 | break; | |
209 | case 'I': | |
210 | err = cn_printf(cn, "%d", | |
211 | task_pid_nr(current)); | |
212 | break; | |
10c28d93 AK |
213 | /* uid */ |
214 | case 'u': | |
5202efe5 NI |
215 | err = cn_printf(cn, "%u", |
216 | from_kuid(&init_user_ns, | |
217 | cred->uid)); | |
10c28d93 AK |
218 | break; |
219 | /* gid */ | |
220 | case 'g': | |
5202efe5 NI |
221 | err = cn_printf(cn, "%u", |
222 | from_kgid(&init_user_ns, | |
223 | cred->gid)); | |
10c28d93 | 224 | break; |
12a2b4b2 ON |
225 | case 'd': |
226 | err = cn_printf(cn, "%d", | |
227 | __get_dumpable(cprm->mm_flags)); | |
228 | break; | |
10c28d93 AK |
229 | /* signal that caused the coredump */ |
230 | case 's': | |
b4176b7c NI |
231 | err = cn_printf(cn, "%d", |
232 | cprm->siginfo->si_signo); | |
10c28d93 AK |
233 | break; |
234 | /* UNIX time of coredump */ | |
235 | case 't': { | |
03927c8a AB |
236 | time64_t time; |
237 | ||
238 | time = ktime_get_real_seconds(); | |
239 | err = cn_printf(cn, "%lld", time); | |
10c28d93 AK |
240 | break; |
241 | } | |
242 | /* hostname */ | |
923bed03 | 243 | case 'h': |
10c28d93 | 244 | down_read(&uts_sem); |
923bed03 | 245 | err = cn_esc_printf(cn, "%s", |
10c28d93 AK |
246 | utsname()->nodename); |
247 | up_read(&uts_sem); | |
10c28d93 | 248 | break; |
10c28d93 | 249 | /* executable */ |
923bed03 ON |
250 | case 'e': |
251 | err = cn_esc_printf(cn, "%s", current->comm); | |
10c28d93 | 252 | break; |
10c28d93 AK |
253 | case 'E': |
254 | err = cn_print_exe_file(cn); | |
255 | break; | |
256 | /* core limit size */ | |
257 | case 'c': | |
258 | err = cn_printf(cn, "%lu", | |
259 | rlimit(RLIMIT_CORE)); | |
260 | break; | |
261 | default: | |
262 | break; | |
263 | } | |
264 | ++pat_ptr; | |
265 | } | |
266 | ||
267 | if (err) | |
268 | return err; | |
269 | } | |
270 | ||
888ffc59 | 271 | out: |
10c28d93 AK |
272 | /* Backward compatibility with core_uses_pid: |
273 | * | |
274 | * If core_pattern does not include a %p (as is the default) | |
275 | * and core_uses_pid is set, then .%pid will be appended to | |
276 | * the filename. Do not do this for piped commands. */ | |
277 | if (!ispipe && !pid_in_pattern && core_uses_pid) { | |
278 | err = cn_printf(cn, ".%d", task_tgid_vnr(current)); | |
279 | if (err) | |
280 | return err; | |
281 | } | |
10c28d93 AK |
282 | return ispipe; |
283 | } | |
284 | ||
5fa534c9 | 285 | static int zap_process(struct task_struct *start, int exit_code, int flags) |
10c28d93 AK |
286 | { |
287 | struct task_struct *t; | |
288 | int nr = 0; | |
289 | ||
5fa534c9 ON |
290 | /* ignore all signals except SIGKILL, see prepare_signal() */ |
291 | start->signal->flags = SIGNAL_GROUP_COREDUMP | flags; | |
10c28d93 AK |
292 | start->signal->group_exit_code = exit_code; |
293 | start->signal->group_stop_count = 0; | |
294 | ||
d61ba589 | 295 | for_each_thread(start, t) { |
10c28d93 AK |
296 | task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK); |
297 | if (t != current && t->mm) { | |
298 | sigaddset(&t->pending.signal, SIGKILL); | |
299 | signal_wake_up(t, 1); | |
300 | nr++; | |
301 | } | |
d61ba589 | 302 | } |
10c28d93 AK |
303 | |
304 | return nr; | |
305 | } | |
306 | ||
403bad72 ON |
307 | static int zap_threads(struct task_struct *tsk, struct mm_struct *mm, |
308 | struct core_state *core_state, int exit_code) | |
10c28d93 AK |
309 | { |
310 | struct task_struct *g, *p; | |
311 | unsigned long flags; | |
312 | int nr = -EAGAIN; | |
313 | ||
314 | spin_lock_irq(&tsk->sighand->siglock); | |
315 | if (!signal_group_exit(tsk->signal)) { | |
316 | mm->core_state = core_state; | |
6cd8f0ac | 317 | tsk->signal->group_exit_task = tsk; |
5fa534c9 | 318 | nr = zap_process(tsk, exit_code, 0); |
403bad72 | 319 | clear_tsk_thread_flag(tsk, TIF_SIGPENDING); |
10c28d93 AK |
320 | } |
321 | spin_unlock_irq(&tsk->sighand->siglock); | |
322 | if (unlikely(nr < 0)) | |
323 | return nr; | |
324 | ||
aed8adb7 | 325 | tsk->flags |= PF_DUMPCORE; |
10c28d93 AK |
326 | if (atomic_read(&mm->mm_users) == nr + 1) |
327 | goto done; | |
328 | /* | |
329 | * We should find and kill all tasks which use this mm, and we should | |
330 | * count them correctly into ->nr_threads. We don't take tasklist | |
331 | * lock, but this is safe wrt: | |
332 | * | |
333 | * fork: | |
334 | * None of sub-threads can fork after zap_process(leader). All | |
335 | * processes which were created before this point should be | |
336 | * visible to zap_threads() because copy_process() adds the new | |
337 | * process to the tail of init_task.tasks list, and lock/unlock | |
338 | * of ->siglock provides a memory barrier. | |
339 | * | |
340 | * do_exit: | |
341 | * The caller holds mm->mmap_sem. This means that the task which | |
342 | * uses this mm can't pass exit_mm(), so it can't exit or clear | |
343 | * its ->mm. | |
344 | * | |
345 | * de_thread: | |
346 | * It does list_replace_rcu(&leader->tasks, ¤t->tasks), | |
347 | * we must see either old or new leader, this does not matter. | |
348 | * However, it can change p->sighand, so lock_task_sighand(p) | |
349 | * must be used. Since p->mm != NULL and we hold ->mmap_sem | |
350 | * it can't fail. | |
351 | * | |
352 | * Note also that "g" can be the old leader with ->mm == NULL | |
353 | * and already unhashed and thus removed from ->thread_group. | |
354 | * This is OK, __unhash_process()->list_del_rcu() does not | |
355 | * clear the ->next pointer, we will find the new leader via | |
356 | * next_thread(). | |
357 | */ | |
358 | rcu_read_lock(); | |
359 | for_each_process(g) { | |
360 | if (g == tsk->group_leader) | |
361 | continue; | |
362 | if (g->flags & PF_KTHREAD) | |
363 | continue; | |
d61ba589 ON |
364 | |
365 | for_each_thread(g, p) { | |
366 | if (unlikely(!p->mm)) | |
367 | continue; | |
368 | if (unlikely(p->mm == mm)) { | |
369 | lock_task_sighand(p, &flags); | |
370 | nr += zap_process(p, exit_code, | |
371 | SIGNAL_GROUP_EXIT); | |
372 | unlock_task_sighand(p, &flags); | |
10c28d93 | 373 | } |
d61ba589 ON |
374 | break; |
375 | } | |
10c28d93 AK |
376 | } |
377 | rcu_read_unlock(); | |
378 | done: | |
379 | atomic_set(&core_state->nr_threads, nr); | |
380 | return nr; | |
381 | } | |
382 | ||
383 | static int coredump_wait(int exit_code, struct core_state *core_state) | |
384 | { | |
385 | struct task_struct *tsk = current; | |
386 | struct mm_struct *mm = tsk->mm; | |
387 | int core_waiters = -EBUSY; | |
388 | ||
389 | init_completion(&core_state->startup); | |
390 | core_state->dumper.task = tsk; | |
391 | core_state->dumper.next = NULL; | |
392 | ||
393 | down_write(&mm->mmap_sem); | |
394 | if (!mm->core_state) | |
395 | core_waiters = zap_threads(tsk, mm, core_state, exit_code); | |
396 | up_write(&mm->mmap_sem); | |
397 | ||
398 | if (core_waiters > 0) { | |
399 | struct core_thread *ptr; | |
400 | ||
401 | wait_for_completion(&core_state->startup); | |
402 | /* | |
403 | * Wait for all the threads to become inactive, so that | |
404 | * all the thread context (extended register state, like | |
405 | * fpu etc) gets copied to the memory. | |
406 | */ | |
407 | ptr = core_state->dumper.next; | |
408 | while (ptr != NULL) { | |
409 | wait_task_inactive(ptr->task, 0); | |
410 | ptr = ptr->next; | |
411 | } | |
412 | } | |
413 | ||
414 | return core_waiters; | |
415 | } | |
416 | ||
acdedd99 | 417 | static void coredump_finish(struct mm_struct *mm, bool core_dumped) |
10c28d93 AK |
418 | { |
419 | struct core_thread *curr, *next; | |
420 | struct task_struct *task; | |
421 | ||
6cd8f0ac | 422 | spin_lock_irq(¤t->sighand->siglock); |
acdedd99 ON |
423 | if (core_dumped && !__fatal_signal_pending(current)) |
424 | current->signal->group_exit_code |= 0x80; | |
6cd8f0ac ON |
425 | current->signal->group_exit_task = NULL; |
426 | current->signal->flags = SIGNAL_GROUP_EXIT; | |
427 | spin_unlock_irq(¤t->sighand->siglock); | |
428 | ||
10c28d93 AK |
429 | next = mm->core_state->dumper.next; |
430 | while ((curr = next) != NULL) { | |
431 | next = curr->next; | |
432 | task = curr->task; | |
433 | /* | |
434 | * see exit_mm(), curr->task must not see | |
435 | * ->task == NULL before we read ->next. | |
436 | */ | |
437 | smp_mb(); | |
438 | curr->task = NULL; | |
439 | wake_up_process(task); | |
440 | } | |
441 | ||
442 | mm->core_state = NULL; | |
443 | } | |
444 | ||
528f827e ON |
445 | static bool dump_interrupted(void) |
446 | { | |
447 | /* | |
448 | * SIGKILL or freezing() interrupt the coredumping. Perhaps we | |
449 | * can do try_to_freeze() and check __fatal_signal_pending(), | |
450 | * but then we need to teach dump_write() to restart and clear | |
451 | * TIF_SIGPENDING. | |
452 | */ | |
453 | return signal_pending(current); | |
454 | } | |
455 | ||
10c28d93 AK |
456 | static void wait_for_dump_helpers(struct file *file) |
457 | { | |
de32ec4c | 458 | struct pipe_inode_info *pipe = file->private_data; |
10c28d93 AK |
459 | |
460 | pipe_lock(pipe); | |
461 | pipe->readers++; | |
462 | pipe->writers--; | |
dc7ee2aa ON |
463 | wake_up_interruptible_sync(&pipe->wait); |
464 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); | |
465 | pipe_unlock(pipe); | |
10c28d93 | 466 | |
dc7ee2aa ON |
467 | /* |
468 | * We actually want wait_event_freezable() but then we need | |
469 | * to clear TIF_SIGPENDING and improve dump_interrupted(). | |
470 | */ | |
471 | wait_event_interruptible(pipe->wait, pipe->readers == 1); | |
10c28d93 | 472 | |
dc7ee2aa | 473 | pipe_lock(pipe); |
10c28d93 AK |
474 | pipe->readers--; |
475 | pipe->writers++; | |
476 | pipe_unlock(pipe); | |
10c28d93 AK |
477 | } |
478 | ||
479 | /* | |
480 | * umh_pipe_setup | |
481 | * helper function to customize the process used | |
482 | * to collect the core in userspace. Specifically | |
483 | * it sets up a pipe and installs it as fd 0 (stdin) | |
484 | * for the process. Returns 0 on success, or | |
485 | * PTR_ERR on failure. | |
486 | * Note that it also sets the core limit to 1. This | |
487 | * is a special value that we use to trap recursive | |
488 | * core dumps | |
489 | */ | |
490 | static int umh_pipe_setup(struct subprocess_info *info, struct cred *new) | |
491 | { | |
492 | struct file *files[2]; | |
493 | struct coredump_params *cp = (struct coredump_params *)info->data; | |
494 | int err = create_pipe_files(files, 0); | |
495 | if (err) | |
496 | return err; | |
497 | ||
498 | cp->file = files[1]; | |
499 | ||
45525b26 AV |
500 | err = replace_fd(0, files[0], 0); |
501 | fput(files[0]); | |
10c28d93 AK |
502 | /* and disallow core files too */ |
503 | current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1}; | |
504 | ||
45525b26 | 505 | return err; |
10c28d93 AK |
506 | } |
507 | ||
ec57941e | 508 | void do_coredump(const siginfo_t *siginfo) |
10c28d93 AK |
509 | { |
510 | struct core_state core_state; | |
511 | struct core_name cn; | |
512 | struct mm_struct *mm = current->mm; | |
513 | struct linux_binfmt * binfmt; | |
514 | const struct cred *old_cred; | |
515 | struct cred *cred; | |
516 | int retval = 0; | |
10c28d93 AK |
517 | int ispipe; |
518 | struct files_struct *displaced; | |
fbb18169 JH |
519 | /* require nonrelative corefile path and be extra careful */ |
520 | bool need_suid_safe = false; | |
acdedd99 | 521 | bool core_dumped = false; |
10c28d93 AK |
522 | static atomic_t core_dump_count = ATOMIC_INIT(0); |
523 | struct coredump_params cprm = { | |
5ab1c309 | 524 | .siginfo = siginfo, |
541880d9 | 525 | .regs = signal_pt_regs(), |
10c28d93 AK |
526 | .limit = rlimit(RLIMIT_CORE), |
527 | /* | |
528 | * We must use the same mm->flags while dumping core to avoid | |
529 | * inconsistency of bit flags, since this flag is not protected | |
530 | * by any locks. | |
531 | */ | |
532 | .mm_flags = mm->flags, | |
533 | }; | |
534 | ||
5ab1c309 | 535 | audit_core_dumps(siginfo->si_signo); |
10c28d93 AK |
536 | |
537 | binfmt = mm->binfmt; | |
538 | if (!binfmt || !binfmt->core_dump) | |
539 | goto fail; | |
540 | if (!__get_dumpable(cprm.mm_flags)) | |
541 | goto fail; | |
542 | ||
543 | cred = prepare_creds(); | |
544 | if (!cred) | |
545 | goto fail; | |
546 | /* | |
547 | * We cannot trust fsuid as being the "true" uid of the process | |
548 | * nor do we know its entire history. We only know it was tainted | |
549 | * so we dump it as root in mode 2, and only into a controlled | |
550 | * environment (pipe handler or fully qualified path). | |
551 | */ | |
e579d2c2 | 552 | if (__get_dumpable(cprm.mm_flags) == SUID_DUMP_ROOT) { |
10c28d93 | 553 | /* Setuid core dump mode */ |
10c28d93 | 554 | cred->fsuid = GLOBAL_ROOT_UID; /* Dump root private */ |
fbb18169 | 555 | need_suid_safe = true; |
10c28d93 AK |
556 | } |
557 | ||
5ab1c309 | 558 | retval = coredump_wait(siginfo->si_signo, &core_state); |
10c28d93 AK |
559 | if (retval < 0) |
560 | goto fail_creds; | |
561 | ||
562 | old_cred = override_creds(cred); | |
563 | ||
12a2b4b2 | 564 | ispipe = format_corename(&cn, &cprm); |
10c28d93 | 565 | |
fb96c475 | 566 | if (ispipe) { |
10c28d93 AK |
567 | int dump_count; |
568 | char **helper_argv; | |
907ed132 | 569 | struct subprocess_info *sub_info; |
10c28d93 AK |
570 | |
571 | if (ispipe < 0) { | |
572 | printk(KERN_WARNING "format_corename failed\n"); | |
573 | printk(KERN_WARNING "Aborting core\n"); | |
e7fd1549 | 574 | goto fail_unlock; |
10c28d93 AK |
575 | } |
576 | ||
577 | if (cprm.limit == 1) { | |
578 | /* See umh_pipe_setup() which sets RLIMIT_CORE = 1. | |
579 | * | |
580 | * Normally core limits are irrelevant to pipes, since | |
581 | * we're not writing to the file system, but we use | |
fcbc32bc | 582 | * cprm.limit of 1 here as a special value, this is a |
10c28d93 AK |
583 | * consistent way to catch recursive crashes. |
584 | * We can still crash if the core_pattern binary sets | |
585 | * RLIM_CORE = !1, but it runs as root, and can do | |
586 | * lots of stupid things. | |
587 | * | |
588 | * Note that we use task_tgid_vnr here to grab the pid | |
589 | * of the process group leader. That way we get the | |
590 | * right pid if a thread in a multi-threaded | |
591 | * core_pattern process dies. | |
592 | */ | |
593 | printk(KERN_WARNING | |
594 | "Process %d(%s) has RLIMIT_CORE set to 1\n", | |
595 | task_tgid_vnr(current), current->comm); | |
596 | printk(KERN_WARNING "Aborting core\n"); | |
597 | goto fail_unlock; | |
598 | } | |
599 | cprm.limit = RLIM_INFINITY; | |
600 | ||
601 | dump_count = atomic_inc_return(&core_dump_count); | |
602 | if (core_pipe_limit && (core_pipe_limit < dump_count)) { | |
603 | printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n", | |
604 | task_tgid_vnr(current), current->comm); | |
605 | printk(KERN_WARNING "Skipping core dump\n"); | |
606 | goto fail_dropcount; | |
607 | } | |
608 | ||
888ffc59 | 609 | helper_argv = argv_split(GFP_KERNEL, cn.corename, NULL); |
10c28d93 AK |
610 | if (!helper_argv) { |
611 | printk(KERN_WARNING "%s failed to allocate memory\n", | |
612 | __func__); | |
613 | goto fail_dropcount; | |
614 | } | |
615 | ||
907ed132 LDM |
616 | retval = -ENOMEM; |
617 | sub_info = call_usermodehelper_setup(helper_argv[0], | |
618 | helper_argv, NULL, GFP_KERNEL, | |
619 | umh_pipe_setup, NULL, &cprm); | |
620 | if (sub_info) | |
621 | retval = call_usermodehelper_exec(sub_info, | |
622 | UMH_WAIT_EXEC); | |
623 | ||
10c28d93 AK |
624 | argv_free(helper_argv); |
625 | if (retval) { | |
888ffc59 | 626 | printk(KERN_INFO "Core dump to |%s pipe failed\n", |
10c28d93 AK |
627 | cn.corename); |
628 | goto close_fail; | |
fb96c475 | 629 | } |
10c28d93 AK |
630 | } else { |
631 | struct inode *inode; | |
632 | ||
633 | if (cprm.limit < binfmt->min_coredump) | |
634 | goto fail_unlock; | |
635 | ||
fbb18169 | 636 | if (need_suid_safe && cn.corename[0] != '/') { |
10c28d93 AK |
637 | printk(KERN_WARNING "Pid %d(%s) can only dump core "\ |
638 | "to fully qualified path!\n", | |
639 | task_tgid_vnr(current), current->comm); | |
640 | printk(KERN_WARNING "Skipping core dump\n"); | |
641 | goto fail_unlock; | |
642 | } | |
643 | ||
fbb18169 JH |
644 | /* |
645 | * Unlink the file if it exists unless this is a SUID | |
646 | * binary - in that case, we're running around with root | |
647 | * privs and don't want to unlink another user's coredump. | |
648 | */ | |
649 | if (!need_suid_safe) { | |
650 | mm_segment_t old_fs; | |
651 | ||
652 | old_fs = get_fs(); | |
653 | set_fs(KERNEL_DS); | |
654 | /* | |
655 | * If it doesn't exist, that's fine. If there's some | |
656 | * other problem, we'll catch it at the filp_open(). | |
657 | */ | |
658 | (void) sys_unlink((const char __user *)cn.corename); | |
659 | set_fs(old_fs); | |
660 | } | |
661 | ||
662 | /* | |
663 | * There is a race between unlinking and creating the | |
664 | * file, but if that causes an EEXIST here, that's | |
665 | * fine - another process raced with us while creating | |
666 | * the corefile, and the other process won. To userspace, | |
667 | * what matters is that at least one of the two processes | |
668 | * writes its coredump successfully, not which one. | |
669 | */ | |
10c28d93 | 670 | cprm.file = filp_open(cn.corename, |
fbb18169 JH |
671 | O_CREAT | 2 | O_NOFOLLOW | |
672 | O_LARGEFILE | O_EXCL, | |
10c28d93 AK |
673 | 0600); |
674 | if (IS_ERR(cprm.file)) | |
675 | goto fail_unlock; | |
676 | ||
496ad9aa | 677 | inode = file_inode(cprm.file); |
10c28d93 AK |
678 | if (inode->i_nlink > 1) |
679 | goto close_fail; | |
680 | if (d_unhashed(cprm.file->f_path.dentry)) | |
681 | goto close_fail; | |
682 | /* | |
683 | * AK: actually i see no reason to not allow this for named | |
684 | * pipes etc, but keep the previous behaviour for now. | |
685 | */ | |
686 | if (!S_ISREG(inode->i_mode)) | |
687 | goto close_fail; | |
688 | /* | |
40f705a7 JH |
689 | * Don't dump core if the filesystem changed owner or mode |
690 | * of the file during file creation. This is an issue when | |
691 | * a process dumps core while its cwd is e.g. on a vfat | |
692 | * filesystem. | |
10c28d93 AK |
693 | */ |
694 | if (!uid_eq(inode->i_uid, current_fsuid())) | |
695 | goto close_fail; | |
40f705a7 JH |
696 | if ((inode->i_mode & 0677) != 0600) |
697 | goto close_fail; | |
86cc0584 | 698 | if (!(cprm.file->f_mode & FMODE_CAN_WRITE)) |
10c28d93 AK |
699 | goto close_fail; |
700 | if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file)) | |
701 | goto close_fail; | |
702 | } | |
703 | ||
704 | /* get us an unshared descriptor table; almost always a no-op */ | |
705 | retval = unshare_files(&displaced); | |
706 | if (retval) | |
707 | goto close_fail; | |
708 | if (displaced) | |
709 | put_files_struct(displaced); | |
e86d35c3 AV |
710 | if (!dump_interrupted()) { |
711 | file_start_write(cprm.file); | |
712 | core_dumped = binfmt->core_dump(&cprm); | |
713 | file_end_write(cprm.file); | |
714 | } | |
10c28d93 AK |
715 | if (ispipe && core_pipe_limit) |
716 | wait_for_dump_helpers(cprm.file); | |
717 | close_fail: | |
718 | if (cprm.file) | |
719 | filp_close(cprm.file, NULL); | |
720 | fail_dropcount: | |
721 | if (ispipe) | |
722 | atomic_dec(&core_dump_count); | |
723 | fail_unlock: | |
724 | kfree(cn.corename); | |
acdedd99 | 725 | coredump_finish(mm, core_dumped); |
10c28d93 AK |
726 | revert_creds(old_cred); |
727 | fail_creds: | |
728 | put_cred(cred); | |
729 | fail: | |
730 | return; | |
731 | } | |
732 | ||
733 | /* | |
734 | * Core dumping helper functions. These are the only things you should | |
735 | * do on a core-file: use only these functions to write out all the | |
736 | * necessary info. | |
737 | */ | |
ecc8c772 AV |
738 | int dump_emit(struct coredump_params *cprm, const void *addr, int nr) |
739 | { | |
740 | struct file *file = cprm->file; | |
2507a4fb AV |
741 | loff_t pos = file->f_pos; |
742 | ssize_t n; | |
ecc8c772 AV |
743 | if (cprm->written + nr > cprm->limit) |
744 | return 0; | |
2507a4fb AV |
745 | while (nr) { |
746 | if (dump_interrupted()) | |
747 | return 0; | |
52da40ae | 748 | n = __kernel_write(file, addr, nr, &pos); |
2507a4fb AV |
749 | if (n <= 0) |
750 | return 0; | |
751 | file->f_pos = pos; | |
752 | cprm->written += n; | |
753 | nr -= n; | |
754 | } | |
ecc8c772 AV |
755 | return 1; |
756 | } | |
757 | EXPORT_SYMBOL(dump_emit); | |
758 | ||
9b56d543 | 759 | int dump_skip(struct coredump_params *cprm, size_t nr) |
10c28d93 | 760 | { |
9b56d543 AV |
761 | static char zeroes[PAGE_SIZE]; |
762 | struct file *file = cprm->file; | |
10c28d93 | 763 | if (file->f_op->llseek && file->f_op->llseek != no_llseek) { |
9b56d543 AV |
764 | if (cprm->written + nr > cprm->limit) |
765 | return 0; | |
528f827e | 766 | if (dump_interrupted() || |
9b56d543 | 767 | file->f_op->llseek(file, nr, SEEK_CUR) < 0) |
10c28d93 | 768 | return 0; |
9b56d543 AV |
769 | cprm->written += nr; |
770 | return 1; | |
10c28d93 | 771 | } else { |
9b56d543 AV |
772 | while (nr > PAGE_SIZE) { |
773 | if (!dump_emit(cprm, zeroes, PAGE_SIZE)) | |
774 | return 0; | |
775 | nr -= PAGE_SIZE; | |
10c28d93 | 776 | } |
9b56d543 | 777 | return dump_emit(cprm, zeroes, nr); |
10c28d93 | 778 | } |
10c28d93 | 779 | } |
9b56d543 | 780 | EXPORT_SYMBOL(dump_skip); |
22a8cb82 AV |
781 | |
782 | int dump_align(struct coredump_params *cprm, int align) | |
783 | { | |
784 | unsigned mod = cprm->written & (align - 1); | |
785 | if (align & (align - 1)) | |
db51242d AV |
786 | return 0; |
787 | return mod ? dump_skip(cprm, align - mod) : 1; | |
22a8cb82 AV |
788 | } |
789 | EXPORT_SYMBOL(dump_align); |